The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection, jailbreak attempts, or instructions to bypass safety guidelines was found.
[DATA_EXFILTRATION]: The skill uses fetch and navigator.sendBeacon for standard API communication and error reporting. There is no evidence of unauthorized sensitive data access or exfiltration to untrusted domains. Error reporting is directed to a relative endpoint (/api/errors), which is standard practice.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, such as piping downloads to a shell or using dynamic evaluation functions like eval(), were detected.
[OBFUSCATION]: The content is clear, well-commented, and free of obfuscation techniques like base64 encoding, hidden characters, or homoglyphs.
[PERSISTENCE]: The skill implements an offline queue using localStorage, which is a legitimate mechanism for cross-session request persistence in web applications and does not represent a malicious persistence threat.
[DYNAMIC_EXECUTION]: Code snippets use standard JSON.parse and JSON.stringify for data handling. No unsafe deserialization or runtime code generation patterns were observed.
[INDIRECT_PROMPT_INJECTION]: While the skill handles external API data, it is a library of developer patterns rather than an active agent skill that processes untrusted data into prompt context. The patterns themselves do not introduce unique vulnerabilities to the agent platform.

api-client